One observation I see made regularly by commentators on electronic markets is that liquidity can 'disappear in an instant' and is somehow ephemeral. Quotes, they say, tend to fade as soon as the market notices there's a seller around.
I'm always rather bemused by this observation, because it implies that in more manual markets, liquidity is (or was) more stable. That's certainly not my personal observation, having traded many manual markets and many electronic markets over the course of my career. Let’s take a trip down memory lane…
In the mid 90s I had the good fortune of being a Nasdaq market maker in New York with a large firm. Some of you may believe that Nasdaq was an electronic exchange back then, but I can assure you it wasn't. Nasdaq worked like this:
Market makers would enter their quotes manually onto a screen, usually in 1,000s of shares with $1 -$2 spreads. These quotes were not auto-executable like they are today (except for SOES orders, but that's a different story) but broker-dealers had to honour the quote that was on the screen as soon as they picked up the phone.
If for example I had an order to sell shares in Ericsson, I had to call the broker-dealers that were on the best bid at that time. So if the stock had 16 broker dealers on the bid, each bidding $21.25 for 1,000 shares, I'd call one and say, “10,000 Ericsson at a quarter, selling them”. That broker would say “buy a 1,000” because that was the size he was bidding for and he'd move his quote down by 1/8.
The problem with working at a large firm, like the one I worked for, was that most of the brokers knew our orders tended to be large, so they very often would only honour the size they were bidding for on the screen when they picked up the phone (as I said this was typically 1,000 shares). We were perhaps able to do this two or three times, selling 2,000-3,000 shares in the process, but by the time we called the fourth broker the market had woken up to the fact there was a seller around because the other traders saw quotes being moved down and stock trading on the bid side. By then the remaining 13 brokers would have moved their quote down by at least 1/8. So perhaps one could say that liquidity took longer to disappear than today, but for all practical purposes you couldn't physically reach it in the first place.
To maximize our fill rate, we developed a primitive form of smart order routing. Suppose we wanted to sell 16,000 shares to the 16 brokers bidding $21.25. The best way to do this was to ask our colleagues on the trading desk to help make calls all at the same time. This way we managed to perhaps sell 13,000/14,000 out of the 16,000. To increase our fill rate, we called some of the other large bulge-bracket firms first because they were notoriously slow at picking up phones. Execution-only broker-dealers on the other hand were very quick at answering phones, so we would sequence our calls to call them last. This way we managed to increase the number of shares we could sell before the brokers were able to react by “fading” their quote.
Now let's turn to the current era. I was rather bemused to find out by reading the book Flash Boys that some brokers struggled with exactly the same phenomenon in an electronic market some 15 years later. I was even more surprised they couldn't figure out why quotes were fading when they tried to access them on multiple exchanges sequentially. I guess I had (wrongly) assumed this was rather obvious from my experience in my Nasdaq days. In a fragmented market, liquidity can be notoriously hard to access if you don't time your orders correctly. That's why, back in the mists of time, a Nasdaq trading desk would all make calls at the same time, rather than one trader doing it sequentially. Many routers today operate in a similar way.
Today, this age old phenomenon of quote fading is sometimes referred to as “latency arbitrage” or worse (and entirely inaccurately) “front running”. This is the central claim of the book Flash Boys and some people would argue it is definitive proof that the US stock markets are, in their words, “rigged”. But this phenomenon is by no means new. As with so many areas of life, it’s just faster. The good news is that many people have found better ways to access fragmented liquidity by developing smarter ways to route their orders. Just like we did in the old days.
 This was pre-decimalisation, so quote-increments were in 1/8s (12.5 cents) rather than one cent. Trading in 12.5 cent increments had the advantage of being able to update quotes manually, but missing volume was obviously a lot more expensive. Post decimalisation, updating quotes manually was no longer humanly possible, because the number of quote updates went up perhaps tenfold. This was one of the main reasons behind the emergence of technology enabled market makers in the early 2000s.
 You may ask why these market makers adjust their quotes. They are simply acting as an intermediary between buyers and sellers and in that capacity adjusting their prices to new information on demand/supply characteristics. The new information in this example is clearly that there is a large seller around and the prices should now be lower based on demand/supply. This is perhaps not great for the seller (in this case me) but it makes for a more efficient market (and it certainly is good for any eventual buyers). If they wouldn’t adjust their quotes they’d also be stuck with inventory they couldn’t offload.
 There's no doubt that US Equity markets today are a lot cheaper and more efficient than they were before, but unfortunately they've also become a lot more complex. Some good proposals on simplifying markets can be found here: https://www.fia.org/resources/fia-ptg-simplify-markets-eliminating-trade-through-rule-and-ban-locked-and-crossed